TY - JOUR
T1 - Generalization bottleneck in deep metric learning
AU - Hu, Zhanxuan
AU - Wu, Danyang
AU - Nie, Feiping
AU - Wang, Rong
N1 - Publisher Copyright:
© 2021 Elsevier Inc.
PY - 2021/12
Y1 - 2021/12
N2 - Deep metric learning aims to learn a non-linear function that maps raw-data to a discriminative lower-dimensional embedding space, where semantically similar samples have larger similarity than dissimilar ones. Most existing approaches process each raw-data in two steps, by mapping the raw-data to a higher-dimensional feature space via a fixed backbone, followed by mapping the higher-dimensional feature space to a lower-dimensional embedding space via a linear layer. This paradigm, however, inevitably leads to a Generalization Bottleneck (GB) problem. Specifically, GB refers to a limitation that the generalization capacity of lower-dimensional embedding space is inferior to the higher-dimensional feature space in the test stage. To mitigate the capacity gap between feature space and embedding space, we propose to introduce a fully-learnable module, dubbed Relational Knowledge Preserving (RKP), that improves the generalization capacity of lower-dimensional embedding space by transferring the mutual similarity of instances. Our proposed RKP module can be integrated into a general deep metric learning approach. And, experiments conducted on different benchmarks show that it can significantly improve the performance of original model.
AB - Deep metric learning aims to learn a non-linear function that maps raw-data to a discriminative lower-dimensional embedding space, where semantically similar samples have larger similarity than dissimilar ones. Most existing approaches process each raw-data in two steps, by mapping the raw-data to a higher-dimensional feature space via a fixed backbone, followed by mapping the higher-dimensional feature space to a lower-dimensional embedding space via a linear layer. This paradigm, however, inevitably leads to a Generalization Bottleneck (GB) problem. Specifically, GB refers to a limitation that the generalization capacity of lower-dimensional embedding space is inferior to the higher-dimensional feature space in the test stage. To mitigate the capacity gap between feature space and embedding space, we propose to introduce a fully-learnable module, dubbed Relational Knowledge Preserving (RKP), that improves the generalization capacity of lower-dimensional embedding space by transferring the mutual similarity of instances. Our proposed RKP module can be integrated into a general deep metric learning approach. And, experiments conducted on different benchmarks show that it can significantly improve the performance of original model.
KW - Feature learning
KW - Image retrieval
KW - Low-dimensional embedding
KW - Metric learning
UR - http://www.scopus.com/inward/record.url?scp=85115992021&partnerID=8YFLogxK
U2 - 10.1016/j.ins.2021.09.023
DO - 10.1016/j.ins.2021.09.023
M3 - 文章
AN - SCOPUS:85115992021
SN - 0020-0255
VL - 581
SP - 249
EP - 261
JO - Information Sciences
JF - Information Sciences
ER -